Calculating Rotation Matrix from Yaw/Pitch/Roll (CC SDK)

I have a use case where I need to update the metadata for each photo from a data file that has more accurate position and yaw/pitch/roll data than what is contained in each photo's exifdata.

Through the SDK, I am able to modify each photo's PoseMetadata Point3d `center` and Matrix3 `rotation` attributes. To change `rotation`, I must find the transformation from yaw/pitch/roll to ContextCapture rotation matrix.

However, the rotation matrix formula provided in the CC user guide does not appear to be correct (docs.bentley.com/.../GUID-2D452A8A-A4FE-450D-A0CA-9336DCF1238A.html).

For instance, if I add a photo with 0 yaw/0 pitch/0 roll exifdata (confirmed through CC SDK photo.exifData.yawPitchRoll), the photo's poseMetadata rotation matrix is [[0.9703417, -0.24173759, -0.00000276], [0.20574631, 0.8258776, -0.52497107], [0.12690751, 0.5094007, 0.85112005]], whereas the provided formula results in [[1, -0, 0], [0, 0, -1], [0, 1, 0]].

I have created the following minimum reproduceable example using any image named "example_image_0.jpg" with yaw/pitch/roll exifdata.

import ccmasterkernel
import math
import numpy as np
np.set_printoptions(suppress=True)

def compare_rotation_matrices():
    print("MasterKernel version %s" % ccmasterkernel.version())
    print()

    project = ccmasterkernel.Project()
    block = ccmasterkernel.Block(project)
    project.addBlock(block)
    photogroups = block.getPhotogroups()

    path = "example_image_0.jpg"
    print("----------", path, "----------")
    photo = photogroups.addPhotoInAutoMode(path)

    yaw, pitch, roll = photo.exifData.yawPitchRoll
    print("--- PHOTO EXIFDATA ---")
    print("Yaw:", yaw, "\tPitch:", pitch, "\tRoll:", roll)
    print()

    rot_mat = np.asarray(photo.poseMetadata.rotation.getElements(), dtype=np.float32).reshape((3, 3))
    print("--- CONTEXTCAPTURE ROTATION MATRIX ---")
    print(rot_mat)
    print()

    calculated_rot_mat = yaw_pitch_roll_to_rotation_matrix(yaw=yaw, pitch=pitch, roll=roll)
    print("--- CALCULATED ROTATION MATRIX ---")
    print(calculated_rot_mat)
    print()


def yaw_pitch_roll_to_rotation_matrix(yaw, pitch, roll):
    """
    Formula from ContextCapture User Manual.
    https://docs.bentley.com/LiveContent/web/ContextCapture%20Help-v10/en/GUID-2D452A8A-A4FE-450D-A0CA-9336DCF1238A.html
    """
    P = np.deg2rad(pitch)
    R = np.deg2rad(roll)
    Y = np.deg2rad(yaw)
    M_00 =  math.cos(R) * math.cos(Y) - math.sin(R) * math.sin(P) * math.sin(Y)
    M_01 = -math.cos(R) * math.sin(Y) - math.cos(Y) * math.sin(R) * math.sin(P)
    M_02 =  math.cos(P) * math.sin(R)
    M_10 =  math.cos(Y) * math.sin(R) + math.cos(R) * math.sin(P) * math.sin(Y)
    M_11 =  math.cos(R) * math.cos(Y) * math.sin(P) - math.sin(R) * math.sin(Y)
    M_12 = -math.cos(R) * math.cos(P)
    M_20 = math.cos(P) * math.sin(Y)
    M_21 = math.cos(P) * math.cos(Y)
    M_22 = math.sin(P)
    return np.array([[M_00, M_01, M_02],
                     [M_10, M_11, M_12],
                     [M_20, M_21, M_22]])


if __name__ == "__main__":
    compare_rotation_matrices()

Here is the output using two example images I have:

As can be seen, the matrices are not the same, so I cannot update the photo rotation with confidence.

Please get back to me. Thank you in advance.